影像分割中Dice系数和Hausdorff距离的比较

Comparison of Dice similarity coefficient and Hausdorff distance in image segmentation

目的:比较影像分割技术中Dice系数和Hausdorff距离的评价效果,分析这两种相似性系数之间的联系,并设计相应的类型加以进一步论述。方法:采用Dice系数和Hausdorff距离对两种轮廓相似度进行评估。设计18个(9对)从临床靶区中抽象出的轮廓,计算并比较这两类相似性系数。结果:根据比较结果,总结出3种不同类型情况:(1)图像符合度好;(2)图像整体符合度好,但存在一小部分符合度差;(3)图像轮廓符合度差。结论:仅靠Dice系数作为唯一评判标准是不完善的,应在此基础上增加Hausdorff距离的筛选,这样才能最大限度地说明轮廓之间的符合程度。

Objective To compare the evaluation effects of Dice similarity coefficient and Hausdorff distance in image segmentation;analyze the relationships between two similarity coefficients;and design the corresponding contours for further discussion. Methods Dice similarity coefficient and Hausdorff distance were used to evaluate contour similarity. Moreover, 18(9 pairs) virtual contours abstracted from clinical target areas were designed for calculating and comparing the two similarity coefficients, namely Dice similarity coefficient and Hausdorff distance. Results According to the comparison results, 3 different types of cases were summarized.(1) The image conformity was preferable;(2) the overall conformity of the images was good, but there was a small proportion of poor conformity;(3) the conformity of image contour was poor. Conclusion Using Dice similarity coefficient as the only criterion is imperfect, and Dice similarity coefficient should be combined with Hausdorff distance to reflect contour conformity to the greatest extent.